4046 new failures...

[Geoff Keating]

> Date: Tue, 26 Feb 2002 23:49:15 +0000 (GMT)
> From: "Joseph S. Myers" <jsm28@cam.ac.uk>
> X-X-Sender:  <jsm28@kern.srcf.societies.cam.ac.uk>
> cc: <gcc-regression@gcc.gnu.org>, <mark@codesourcery.com>, <aldyh@redhat.com>,
>    <degger@fhm.edu>, <rth@redhat.com>
> 
> On Tue, 26 Feb 2002, Geoff Keating wrote:
> 
> > This turned out to be because GCC's version number changed, and so it
> > couldn't find the powerpc-eabisim assembler (it was looking for it in
> > lib/gcc-lib/powerpc-eabisim/3.2/../../../../powerpc-eabisim/bin/as
> > but there was no '3.2' directory; I have created one now).
> > 
> > Mark, could it be arranged that there is a ChangeLog entry when the
> > version number changes like that?  It's a significant user-visible
> > change, and it also helps people reading the ChangeLog.
> 
> Could it be arranged, before any future such change with a ChangeLog entry
> (I put one in a year ago when increasing the mainline version then to 3.1,
> and the same problem occurred), that the necessary fixes are made so that
> the person making the change doesn't receive bogus regression reports?

I'm not sure how this bug should be fixed.

There seem to be four obvious choices:

1. Change GCC so that, when searching for files, it simplifies
   filenames so that they don't contain "<file>/..".  The problem is
   that this can cause trouble with symlinks, because sometimes
   "<file>/.." is not the same as "."

2. Change the build process so that such paths don't get generated.
   The problem is that this means that 'make exec_prefix=foo" doesn't
   work right.

3. Change configure so that the appropriate directory is created if it
   doesn't exist.  This has the disadvantage of (2) and the additional
   disadvantage that just configuring will change the install tree.

4. Declare it a feature and try to work around it in the regression
   tester (by having gcc/contrib/regression/btest-gcc.sh create the
   appropriate directory in $PREFIX).  But it's not a very good
   feature.

Can anyone think of a better way?

-- 
- Geoffrey Keating <geoffk@geoffk.org> <geoffk@redhat.com>